PANSURAS

Perioperative ANesthesia & SURgical Assessment System

The How — Observation & Machine Learning

Generalized accurate prediction of postoperative outcomes is almost impossible with random controlled trials, because there are simply too many factors influencing each postoperative outcome. This was eloquently formulated in 1996 by Prof. Nick Black.

… “most interventions have many components. Consider a simple surgical operation: this entails preoperative tests, anesthesia, the surgical approach, wound management, post-operative nursing, and discharge practice. And these are just the principal components. It will only ever be practical to subject a limited number of items to experimental evaluation.”

Furthermore, these factors differ between different institutions within each region, as well as between countries. The only way to solve this problem is to analyze the outcomes per institution — to use observational science (Black 1996).

Machine learning & Observations

PANSURAS solves these problems with a unique standarized observational data collection system, together with a semi-automated machine-learning process for each each individual institution where it is used. This may seem a “black box” labelled “machine-learning” being applied to analyze data from each institution, but the reality is more prosaic. Machine-learning is no more than a collection of algorithms or computer programs to process data in a database according to rules set by the programmer(s). In the case of PANSURAS, these are transparent, readable, and readily understood code blocks for data extraction and multiple forms of statistical analysis.

Startup — Period of initial use

PANSURAS is not a box of magical tricks. Initially it will not provide data relevant to the exact local situation when initially used. Instead, just as with all machine-learning it must first have sufficient data from the intitution where used before any risk surgical risk factors and coefficients for predictive equations can be derived. Therefore users in each institution are faced with an initial and subsequent phases.

  • The PANSURAS interface provides a structured user-friendly interface with which to perform preoperative assessment with very useful on-the-fly decision support and scoring systems to aid decision-making. Hospital information systems are not designed to provide decision support for perioperative processes — they are usually no more than storage and record systems designed primarily to aid organisational processes within hospitals or institutions. This makes initial startup use of PANSURAS very useful in its own right, because it provides much, much more decision-support than all known hospital information systems.
  • Follow-up data entry is equally user-friendly, structured, and able to be done very rapidly, much of it while the patient is still admitted to hospital.
  • Initally, insufficient data for machine-learning is present in institutions where users start to use PANSURAS. This means there are no known coefficients for predictive algorithms specific to that new location. One solution to this problem is to use a set of known coefficients from other institutions from another geographical location. Otherwise, users must wait until sufficient local data is collected.

Continued use & sufficient data

The flow chart below shows how the user interface is an initial algorithmic decison support system based upon scoring systems and studies published in medical journals, while the output for individual patients is a combination of the latter with location-specific machine-learning.


Flow chart of user interaction with PANSURAS. Note the initial feedback with algorithmic data, and the use of the “simulation mode” to determine the effects of measures to aid in preoperative optimization. (Click image to enlarge)

After collection of sufficient data at the location where employed, and activation of the data extraction and statistical modules forming the machine-learning system:

  • PANSURAS extracts the risk factors per postoperative outcome / complication per operation for the location where used.
  • It applies weighting to these risk factors according to the local situation.
  • Generates new coefficients for the predictive equations of each postoperative outcome per operation using these same risk factors.
  • Multiple statistical, quality, benchmarking, and audit facilities present in PANSURAS are now available for those with sufficient user-rights.
  • At the same time, the PANSURAS interface continues to provide a structured user-friendly interface with which to perform preoperative assessment with very useful on-the-fly decision support and scoring systems to aid decision-making.

As a result, after sufficient data is present in the database, PANSURAS generates individualized risk predictions SPECIFIC for that institution to users involved in patient assessment — something truly unique at this time, and making the analytic system of PANSURAS applicable worldwide.

Reference:

Black N. Why we need observational studies to evaluate the effectiveness of health care. British Medical Journal, 1996; 312: 1215-1218.


Home »